Skip to contentSkip to conversation

Thinkpad X200s

PATCH Ian Molton 2 participants 8 messages Nov 27, 2009 – Dec 2, 2009

Subsystems (2)
lore
Share & tools
lore

Ian Molton

author
#0 lore raw

Hi folks,

I've been investigating a problem with rfkill on the X200s from Lenovo. The problem exists in debians 2.6.30-2 kernel and persists right up to -git

I've found that the WWAN card does not behave as expected on suspend / resume.

The problem seems to be the rfkill layer getting out of sync with ACPIs idea of the hardware state.

If the laptop is suspended with WWAN enabled, it will wake up with it disabled, requiring a toggle of the physical WWAN switch or a prod in /sys to get it going again.

I wrote a little patch that "cures" this issue (attached), however it is far from perfect.

For example, even with this patch, when the side switch is used, it overrides the soft setting for the devices completely, so if for example the WWAN was off, then toggling the switch changes its rfkill state from 0 -> 2, and then from 2 -> 1

It *appears* from the code that it should be possible to get the device to power up in the state it was last in, but I cant make mine do it.

I've also noticed that on recent kernels, I get an extra rfkillswitch for hci0 (bluetooth) in addition to the ACPI one. This vanishes if power is toggled using the ACPI rfkill softswitch, however bluetooth continues to work (when enabled).

Has anyone else experienced these issues?

-Ian

drivers/platform/x86/thinkpad_acpi.c +26 -0
index d93108d..4896af5 100644
@@ -4055,14 +4055,37 @@ enum {
 };
 
 #define TPACPI_RFK_WWAN_SW_NAME		"tpacpi_wwan_sw"
+static int laststat;
 
 static void wan_suspend(pm_message_t state)
 {
+	int status;
+	acpi_evalf(hkey_handle, &status, "GWAN", "d");
+	laststat = status;
+        printk("s%d stat.\n", status);
 	/* Try to make sure radio will resume powered off */
 	if (!acpi_evalf(NULL, NULL, "\\WGSV", "qvd",
 		   TP_ACPI_WGSV_PWR_OFF_ON_RESUME))
 		vdbg_printk(TPACPI_DBG_RFKILL,
 			"WWAN power down on resume request failed\n");
+	printk("sus\n");
+	acpi_evalf(hkey_handle, &status, "GWAN", "d");
+        printk("s%d stat.\n", status);
+}
+
+static int wan_resume(pm_message_t state)
+{
+	int status;
+	printk("res\n");
+	acpi_evalf(hkey_handle, &status, "GWAN", "d");
+	printk("r%d stat.\n", status);
+        printk("l%d stat.\n", laststat);
+	if(laststat & TP_ACPI_WANCARD_RADIOSSW) {
+	acpi_evalf(hkey_handle, NULL, "SWAN", "vd", TP_ACPI_WANCARD_RADIOSSW);
+	acpi_evalf(hkey_handle, &status, "GWAN", "d");
+	printk("r%d stat.\n", status);
+	tpacpi_rfk_update_swstate_all();
+	}
 }
 
 static int wan_get_status(void)
@@ -4078,6 +4101,7 @@ static int wan_get_status(void)
 	if (!acpi_evalf(hkey_handle, &status, "GWAN", "d"))
 		return -EIO;
 
+	printk("gs: %d\n", status);
 	return ((status & TP_ACPI_WANCARD_RADIOSSW) != 0) ?
 			TPACPI_RFK_RADIO_ON : TPACPI_RFK_RADIO_OFF;
 }
@@ -4103,6 +4127,7 @@ static int wan_set_status(enum tpacpi_rfkill_state state)
 	else
 		status = 0;
 
+	printk("ss: %d\n", status);
 	if (!acpi_evalf(hkey_handle, NULL, "SWAN", "vd", status))
 		return -EIO;
 
@@ -4239,6 +4264,7 @@ static struct ibm_struct wan_driver_data = {
 	.read = wan_read,
 	.write = wan_write,
 	.exit = wan_exit,
+	.resume = wan_resume,
 	.suspend = wan_suspend,
 	.shutdown = wan_shutdown,
 };
#1 lore raw
Quoted 2 lines · Ian Molton

The problem seems to be the rfkill layer getting out of sync with ACPIs idea of the hardware state.

Hmm... did rfkill drop the suspend/resume support perchance?

Quoted 3 lines

If the laptop is suspended with WWAN enabled, it will wake up with it disabled, requiring a toggle of the physical WWAN switch or a prod in /sys to get it going again.

Yes. rfkill is suposed to do proper suspend/resume in the class level, which would call back thinkpad-acpi and request the radio to be turned on.

Quoted 2 lines

I wrote a little patch that "cures" this issue (attached), however it is far from perfect.

Indeed, and it is NAK'd. Let's fix it properly :-)

Quoted 4 lines

For example, even with this patch, when the side switch is used, it overrides the soft setting for the devices completely, so if for example the WWAN was off, then toggling the switch changes its rfkill state from 0 -> 2, and then from 2 -> 1

...

Quoted 2 lines

It *appears* from the code that it should be possible to get the device to power up in the state it was last in, but I cant make mine do it.

Depends only on rfkill, and whatever is handling EV_SW SW_RFKILL_ALL in your Linux distribution. Sometime ago, you could give rfkill-input a parameter to change its mode to do just what you want. I don't know if that's still possible.

So, at least for this part of the issue, you have to turn your eyes to rfkill and userspace.

Quoted 4 lines

I've also noticed that on recent kernels, I get an extra rfkillswitch for hci0 (bluetooth) in addition to the ACPI one. This vanishes if power is toggled using the ACPI rfkill softswitch, however bluetooth continues to work (when enabled).

Yes. The thinkpad-acpi rfkill switch powers down the bluetooth hardware, thus it disappears.

drivers/platform/x86/thinkpad_acpi.c enum { · line 4088

Quoted 13 lines
drivers/platform/x86/thinkpad_acpi.c +13 -0
+static int wan_resume(pm_message_t state)
+{
+	int status;
+	printk("res\n");
+	acpi_evalf(hkey_handle, &status, "GWAN", "d");
+	printk("r%d stat.\n", status);
+        printk("l%d stat.\n", laststat);
+	if(laststat & TP_ACPI_WANCARD_RADIOSSW) {
+	acpi_evalf(hkey_handle, NULL, "SWAN", "vd", TP_ACPI_WANCARD_RADIOSSW);
+	acpi_evalf(hkey_handle, &status, "GWAN", "d");
+	printk("r%d stat.\n", status);
+	tpacpi_rfk_update_swstate_all();
+	}

This is the right idea, done wrong. I will fix it properly. Just let me check if it needs fixing in thinkpad-acpi or in rfkill, because if it is in rfkill, they might have broken other drivers.

Signature · 4 lines
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh
#2 lore raw
Quoted 5 lines · Henrique de Moraes Holschuh
Quoted 2 lines · Ian Molton

The problem seems to be the rfkill layer getting out of sync with ACPIs idea of the hardware state.

Hmm... did rfkill drop the suspend/resume support perchance?

Yes, for persistent devices, which are thinkpad-acpi and eeepc-laptop. I even ack'd the change that caused this regression, because I didn't pay enough attention to the last lines of the patch. ARGH!

This can be worked-around by thinkpad-acpi with a few lines of extra code, or fixed in rfkill removing two lines of code, which would allow us to remove more LOC from eeepc-laptop as well, as they'd become redundant.

It is a regression from 2.6.30, so it will have to go to stable anyway. This means it is best to waste a few days checking with the others if it would be better fixed in rfkill or thinkpad-acpi, and then get it in 2.6.32 and 2.6.31 through stable.

I will handle it, either way.

Signature · 4 lines
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh
author
#3 lore raw
Quoted 6 lines · Henrique de Moraes Holschuh

It is a regression from 2.6.30, so it will have to go to stable anyway. This means it is best to waste a few days checking with the others if it would be better fixed in rfkill or thinkpad-acpi, and then get it in 2.6.32 and 2.6.31 through stable.

I will handle it, either way.

Cheers - I'll keep an eye on this thread so I can test your patch ASAP! :-)

-Ian

#4 lore raw
Quoted 9 lines · Ian Molton
Quoted 6 lines · Henrique de Moraes Holschuh

It is a regression from 2.6.30, so it will have to go to stable anyway. This means it is best to waste a few days checking with the others if it would be better fixed in rfkill or thinkpad-acpi, and then get it in 2.6.32 and 2.6.31 through stable.

I will handle it, either way.

Cheers - I'll keep an eye on this thread so I can test your patch ASAP! :-)

And I will _need_ your help. Here is a tentative patch. I can use a much heavier hand than this, but if the firmware cooperates, this one should do it (and it will be faster, to boot).

Please check if WWAN, bluetooth and if at all possible, UWB are doing the right thing:

remain enabled if you suspend with them enabled. remain disabled if you suspend with they disabled.

And in the case of WWAN and bluetooth, whether they retain their state (enabled or disabled) across reboots and shutdown.

If you happen to find out that UWB also retains state across reboot/shutdown, I'd like to know about it.

Signature · 4 lines
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh

commit 03dd8d5357d8fe2d91bff4ccdecd7e26f73e92e9

Author: Henrique de Moraes Holschuh <[email protected]>
Date:   Sat Nov 28 00:31:47 2009 -0200

thinkpad-acpi: preserve rfkill state across suspend/resume

Since the rfkill rework in 2.6.31, the driver is always resuming with the radios all in blocked state. We used to resume with radios blocked, and then depend on the rfkill core to re-enable them.

This doesn't work with the new rfkill core and thinkpad-acpi rfkill code for various reasons.

Change thinkpad-acpi to ask the firmware to resume with the radios in the last state when possible, and preserve that state in the driver itself when that's not possible.

This fixes a regression from 2.6.30.

drivers/platform/x86/thinkpad_acpi.c +5 -27
index 915e054..4abde23 100644
@@ -3877,15 +3877,6 @@ enum {
 
 #define TPACPI_RFK_BLUETOOTH_SW_NAME	"tpacpi_bluetooth_sw"
 
-static void bluetooth_suspend(pm_message_t state)
-{
-	/* Try to make sure radio will resume powered off */
-	if (!acpi_evalf(NULL, NULL, "\\BLTH", "vd",
-		   TP_ACPI_BLTH_PWR_OFF_ON_RESUME))
-		vdbg_printk(TPACPI_DBG_RFKILL,
-			"bluetooth power down on resume request failed\n");
-}
-
 static int bluetooth_get_status(void)
 {
 	int status;
@@ -3919,10 +3910,9 @@ static int bluetooth_set_status(enum tpacpi_rfkill_state state)
 #endif
 
 	/* We make sure to keep TP_ACPI_BLUETOOTH_RESUMECTRL off */
+	status = TP_ACPI_BLUETOOTH_RESUMECTRL;
 	if (state == TPACPI_RFK_RADIO_ON)
-		status = TP_ACPI_BLUETOOTH_RADIOSSW;
-	else
-		status = 0;
+		status |= TP_ACPI_BLUETOOTH_RADIOSSW;
 
 	if (!acpi_evalf(hkey_handle, NULL, "SBDC", "vd", status))
 		return -EIO;
@@ -4061,7 +4051,6 @@ static struct ibm_struct bluetooth_driver_data = {
 	.read = bluetooth_read,
 	.write = bluetooth_write,
 	.exit = bluetooth_exit,
-	.suspend = bluetooth_suspend,
 	.shutdown = bluetooth_shutdown,
 };
 
@@ -4079,15 +4068,6 @@ enum {
 
 #define TPACPI_RFK_WWAN_SW_NAME		"tpacpi_wwan_sw"
 
-static void wan_suspend(pm_message_t state)
-{
-	/* Try to make sure radio will resume powered off */
-	if (!acpi_evalf(NULL, NULL, "\\WGSV", "qvd",
-		   TP_ACPI_WGSV_PWR_OFF_ON_RESUME))
-		vdbg_printk(TPACPI_DBG_RFKILL,
-			"WWAN power down on resume request failed\n");
-}
-
 static int wan_get_status(void)
 {
 	int status;
@@ -4120,11 +4100,10 @@ static int wan_set_status(enum tpacpi_rfkill_state state)
 	}
 #endif
 
-	/* We make sure to keep TP_ACPI_WANCARD_RESUMECTRL off */
+	/* We make sure to set TP_ACPI_WANCARD_RESUMECTRL */
+	status = TP_ACPI_WANCARD_RESUMECTRL;
 	if (state == TPACPI_RFK_RADIO_ON)
-		status = TP_ACPI_WANCARD_RADIOSSW;
-	else
-		status = 0;
+		status |= TP_ACPI_WANCARD_RADIOSSW;
 
 	if (!acpi_evalf(hkey_handle, NULL, "SWAN", "vd", status))
 		return -EIO;
@@ -4262,7 +4241,6 @@ static struct ibm_struct wan_driver_data = {
 	.read = wan_read,
 	.write = wan_write,
 	.exit = wan_exit,
-	.suspend = wan_suspend,
 	.shutdown = wan_shutdown,
 };
 
author
#5 lore raw
Quoted 10 lines · Henrique de Moraes Holschuh
Quoted 1 line

Cheers - I'll keep an eye on this thread so I can test your patch ASAP! :-)

And I will _need_ your help. Here is a tentative patch. I can use a much heavier hand than this, but if the firmware cooperates, this one should do it (and it will be faster, to boot).

Please check if WWAN, bluetooth and if at all possible, UWB are doing the right thing:

remain enabled if you suspend with them enabled.

Yes, and:

Quoted 1 line

remain disabled if you suspend with they disabled.

Yes (bluetooth and WWAN)

*BUT* If I toggle the hard RF killswitch, they all forget themselves and reset to 'on' rather than their last known state. IOW, they are disabled, then unconditionally re-enabled.

Quoted 2 lines

And in the case of WWAN and bluetooth, whether they retain their state (enabled or disabled) across reboots and shutdown.

Yes - although GNOME interferes with that later...

Quoted 2 lines

If you happen to find out that UWB also retains state across reboot/shutdown, I'd like to know about it.

I know nothing about the X200s UWB - what module would I need to test?

Hope this helps,

-Ian

#6 lore raw
Quoted 8 lines · Ian Molton
Quoted 1 line · Henrique de Moraes Holschuh

remain disabled if you suspend with they disabled.

Yes (bluetooth and WWAN)

*BUT* If I toggle the hard RF killswitch, they all forget themselves and reset to 'on' rather than their last known state. IOW, they are disabled, then unconditionally re-enabled.

This might well be userspace or the rfkill core, if the EC queues any events and deliver them when the thinkad wakes up, they would enable the radios in the default rfkill config.

Unfortunately, that's not something I can test on my T43. Still, even if it is the firmware doing that, I think we could live with it.

That said, please play with the master_switch_mode parameter of the rfkill module. I personally cannot tolerate anything but master_switch_mode=1, and I dare think it might solve your problem :)

I think I can consider this patch approved and add your tested-by?

Quoted 4 lines
Quoted 2 lines

If you happen to find out that UWB also retains state across reboot/shutdown, I'd like to know about it.

I know nothing about the X200s UWB - what module would I need to test?

I don't know if your x200s has an UWB radio. Does thinkpad-acpi tells you it exists? There is a CONFIG_UWB option which enables UWB and WUSB support.

Signature · 4 lines
  "One disk to rule them all, One disk to find them. One disk to bring
  them all and in the darkness grind them. In the Land of Redmond
  where the shadows lie." -- The Silicon Valley Tarot
  Henrique Holschuh
author Tested-by
#7 lore raw
Quoted 3 lines · Henrique de Moraes Holschuh

That said, please play with the master_switch_mode parameter of the rfkill module. I personally cannot tolerate anything but master_switch_mode=1, and I dare think it might solve your problem :)

Sadly not. It looks like it should, and I've instrumented the rfkill module to be sure the parameter is set, which it is. Debug log (WWAN) follows. Bluetooth behaves the same way as WWAN - powers up after killswitch toggle, regardless of its state prior to that.

# cat /sys/module/rfkill/parameters/master_switch_mode
1

prior to sleep:

# find /sys | grep rfkill..name | xargs cat
tpacpi_bluetooth_sw
tpacpi_wwan_sw
hci0
phy0
# find /sys | grep rfkill..state | xargs cat
1
1
1
1

Turn off WWAN vis /sys:

# echo 0 > /sys/devices/platform/thinkpad_acpi/rfkill/rfkill1/state

Verify it is off:

# find /sys | grep rfkill..state | xargs cat
1
0
1
1

Flip rfkill (hard) switch to off:

corwyn:/home/collabora# find /sys | grep rfkill..state | xargs cat
2
2
2

note: hci0 goes missing when the hard switch is used. (presumably usb power loss)

Flip rfkill (hard) switch back on again:

corwyn:/home/collabora# find /sys | grep rfkill..state | xargs cat
1
1
1
1
Quoted 1 line

I think I can consider this patch approved and add your tested-by?

You can certainly add my tested-by for the sleep/resume path patch.

Quoted 1 line

Does thinkpad-acpi tells

acpi -V -s does not list any wireless devices at all. dmesg has a few mentions of thinkpad-acpi but only the rfkillswitch as far as wireless devices.

Thanks for your quick response too :-)

-Ian